Framework for Customized, Machine Learning Driven Condition Monitoring System for Manufacturing

Abstract Knowledge about the technical state of a complex machine is crucial regarding the reliability and availability within the entire usage phase of a certain technical product. By gathering and gaining information about the exact health condition, production stops can be minimized while optimizing the efficiency of a machine leading to increased customer satisfaction. Since more and more data is available and gets recorded by plenty of sensors, classical statistical methods are not applicable anymore or are just not designed to process such large amount of data. Due to that purpose, many methods from the field of machine learning emerged in the past years. Machine learning in general addresses the issue of recognizing patterns and rules within large amounts of data with the goal to make predictions. Within this paper, the focus is laid upon a detailed step-by-step guide on the development of customized condition monitoring solutions in manufacturing since the application of these still poses a major problem for lots of users. The entire concept of such solutions starting with data collection, through signal-interpretation and index development, finishing with reliable monitoring of the machine state during the manufacturing process is presented throughout. This is done on a real-life application of the concept using synthesized yet realistic data. Moreover, an in-depth analysis of advantages, possibilities and opportunities of the presented concept as well as its weaknesses and boundaries is performed.